[Transfer calibration for alcohol determination using temperature-induced shortwave near infrared spectra].

The authors studied the temperature influence on short-wave near-infrared spectra of ethanol aquatic solution and utilized four methods to establish the transfer partial least squares(PLS) calibration model: direct transfer calibration,global calibration,orthogonal signal correction(OSC) and generalized least squares weighting(GLSW).The PLS models were built at four temperatures: 15,25,35 and 40 ℃.The results showed that direct calibration provided high prediction bias: significantly high positive prediction bias for a temperature lower than calibration temperature and negative bias for higher temperatures.By using the global correction,OSC and GLSW,the systematic errors could be reduced.However,the global correction needed more calibration samples and built a more complex model.The OSC and GLSW methods provided better predictions using fewer latent variables.By using the GLSW method,prediction bias less than 0.1% and RMSEP less than 0.9% were obtained.The absolute prediction error of GLSW method was less than 1.5%.Additionally,the GLSW provided smaller prediction error at every researched temperature using fewer latent variables than OSC.Thus,GLSW was superior to OSC and could establish more robust transfer calibration model.